Developing an Intelligent Waste Classification System Based on Hybrid Deep Transfer Learning Model

Abstract

Background:

The world is more and more interested in sustainable waste management, for which intelligent systems performing automatic sorting of different kinds of waste are needed.

Materials and Methods:

In this study, the hybrid deep learning model is composed of transfer learning with the DenseNet121 model, convolutional block attention module (CBAM) and a simple recurrent neural network (Simple RNN) to enhance feature representation and sequential dependencies of waste images. In order to handle the class imbalance and enrich the data set, preprocessing techniques like normalization, murder and brightness adjusted and strategic data bird mooding (rotating, flipping) were performed.

Results:

The experimental results on a multi-class waste dataset with twelve categories, namely batteries, biowaste, glass types, cardboard, clothing, metals, paper, plastics, shoes, regular waste, and white glass, demonstrated that the model can yield good performance, showing well-balanced classification across all classes. This model reduces the training and parameter size remarkably by using pre-trained weights, and the proposed CBAM mechanism concentrates on learning significant features. The proposed hybrid model achieved an accuracy of 99%, outperforming conventional baselines.

Conclusion:

The experimental results demonstrate the potential to apply the proposed approach in real-world smart recycling systems, which is an effective way to realize waste sorting and environmental protection. Future work will investigate generalization of this technique to other environmental areas, and validate its performance on a larger and more diverse dataset.